GaLactic and Extragalactic All-sky Murchison Widefield Array (GLEAM) survey III: South Galactic Pole data release
T. Franzen, N. Hurley-Walker, S. White, P. Hancock, N. Seymour, A. Kapi?ska, L. Staveley-Smith, R. Wayth
PPublications of the Astronomical Society of Australia (PASA)doi: 10.1017/pas.2021.xxx.
GaLactic and Extragalactic All-sky MurchisonWidefield Array (GLEAM) survey III: South GalacticPole data release
T. M. O. Franzen , , N. Hurley-Walker , S. V. White , , P. J. Hancock , N. Seymour ,A. D. Kapińska , , L. Staveley-Smith , and R. B. Wayth , International Centre for Radio Astronomy Research, Curtin University, Bentley, WA 6102, Australia ASTRON: the Netherlands Institute for Radio Astronomy, PO Box 2, 7990 AA, Dwingeloo, The Netherlands Department of Physics and Electronics, Rhodes University, PO Box 94, Grahamstown, 6140, South Africa International Centre for Radio Astronomy Research, University of Western Australia, Crawley 6009, Australia National Radio Astronomy Observatory, 1003 Lopezville Rd, Socorro, NM 87801, USA ARC Centre of Excellence for All Sky Astrophysics in 3 Dimensions (ASTRO 3D), Australia
Abstract
We present the South Galactic Pole (SGP) data release from the GaLactic and Extragalactic All-skyMurchison Widefield Array (GLEAM) survey. These data combine both years of GLEAM observationsat 72–231 MHz conducted with the Murchison Widefield Array (MWA) and cover an area of 5,113 deg centred on the SGP at 20 h m < RA < h m and − ◦ < Dec < − ◦ . At 216 MHz, the typicalrms noise is ≈ ≈ σ , of which 77 per cent have measured spectral indices between72 and 231 MHz. Improvements to the data reduction in this release include the use of the GLEAMExtragalactic catalogue as a sky model to calibrate the data, a more efficient and automated algorithmto deconvolve the snapshot images, and a more accurate primary beam model to correct the flux scale.This data release enables more sensitive large-scale studies of extragalactic source populations as wellas spectral variability studies on a one-year timescale. Keywords: radio continuum: galaxies — surveys — catalogues — methods: data analysis — techniques:interferometric
Over the past decade, a number of new low-frequency( (cid:46)
200 MHz) radio telescopes have come online, includ-ing the Low Frequency Array (LOFAR; van Haarlemet al., 2013), the Precision Array for Probing the Epochof Reionisation (PAPER; Parsons et al., 2010), the LongWavelength Array (LWA; Ellingson et al., 2013) and theMurchison Widefield Array (MWA; Tingay et al., 2013),as part of preparations for the low-frequency SquareKilometre Array (SKA1 LOW). The design of thesetelescopes was largely guided by the goal of measuringthe redshifted 21 cm signal from the Epoch of Reionisa-tion (EoR), predicted to lie between 50 and 200 MHz(e.g. Furlanetto et al., 2006). Other science objectivesof these telescopes include exploring the transient radiosky, performing Galactic and extragalactic surveys, andtracking solar, heliospheric and ionospheric phenomena. ∗ Email: [email protected]
The large-area sky surveys performed by these instru-ments allow the statistical properties of large samples ofradio galaxies to be investigated, thereby contributing toour understanding of active galactic nucleus (AGN) andstar formation activity over cosmic time (e.g. Simpson,2017). Galaxies across the Universe, including our ownMilky Way, produce strong foreground contamination inexperiments seeking to detect the EoR. Accurate charac-terisation and removal of these foreground contaminantsis a critical step in the interpretation of EoR data (e.g.Procopio et al., 2017; Trott et al., 2019).The GaLactic and Extragalactic All-sky MurchisonWidefield Array (GLEAM; Wayth et al., 2015) surveyis an all-sky continuum survey at 72–231 MHz with anangular resolution of ≈ a r X i v : . [ a s t r o - ph . GA ] F e b Franzen et al. and is a low-frequency precursor telescope for the SKA.The wide range of science enabled by GLEAM, much ofwhich is dependent on the survey’s wide areal coverage,large fractional bandwidth and high surface brightnesssensitivity, is detailed in Beardsley et al. (2019).The GLEAM Extragalactic data release (Exgal;Hurley-Walker et al., 2017) is based on the first year(2013–2014) of GLEAM observations. It covers the en-tire sky south of declination +30 ◦ excluding the strip atGalactic latitude | b | < ◦ , and a few regions affected bypoor ionospheric conditions and around bright, complexsources, such as the Magellanic Clouds. The GLEAMExgal catalogue contains 307,455 component sourcesabove a 5 σ detection limit of ≈
50 mJy / beam, the vastmajority of which have measured in-band spectral in-dices. The sensitivity in GLEAM Exgal is limited bysidelobe confusion, i.e. noise introduced into the imagedue to the combined sidelobes of undeconvolved sources(Franzen et al., 2019), while the flux density calibrationis limited by errors in the primary beam model of order5–20 per cent.The GLEAM Exgal catalogue has been combined withhigher-frequency radio catalogues to measure broad-band spectral energy distributions, enabling detailedstudies of radio-loud AGN (Herzog et al., 2016; Call-ingham et al., 2017) and local powerful star-forminggalaxies (Kapińska et al., 2017; Galvin et al., 2018). Theextremely high surface brightness sensitivity of GLEAMExgal has made possible the detection and characterisa-tion of large, faint objects such as dying radio galaxies(Duchesne & Johnston-Hollitt, 2019), and radio haloesand relics associated with galaxy clusters (Schellenbergeret al., 2017). Franzen et al. (2019) used the GLEAMExgal catalogue to derive low-frequency source countsabove ∼
100 mJy to high precision, allowing tight con-straints on bright radio source population models. Morerecently, White et al. (2020a,b) constructed a completesample of the ‘brightest’ radio sources ( S
151 MHz > ◦ from the GLEAM Exgal cat-alogue, the majority of which are AGN with powerfulradio jets; the GLEAM 4-Jy Sample is an order of magni-tude larger than the 3CRR sample by Laing et al. (1983)and will be a benchmark for the bright radio galaxypopulation.In this paper, we present an extension to GLEAM inorder to reach a sensitivity of ≈ / beam in a ≈ area of sky centred on the South GalacticPole (SGP) at 20 h m < RA < h m and − ◦ < Dec < − ◦ . The GLEAM SGP data release is basedon a subset of the data from both years (2013–2015)of GLEAM observations. The region of sky covered inthis release has been the target of a number of deepmulti-wavelength surveys, such as the Galaxy and MassAssembly (GAMA; Driver et al., 2009) 02 and 23 fields,the Chandra
Deep Field South (CDFS; Giacconi et al.,2001) and the European Large Area ISO Survey - South 1 (ELAIS-S1; Oliver et al., 2000). MWA observationsdedicated to detect the first global signals from theEoR concentrate on two fields, named EoR0 (centred at00 h − ◦ ) and EoR1 (centred at 04 h − ◦ ), which alsolie in this region of sky (Beardsley et al., 2016).We analyse the GLEAM SGP data using an improveddata reduction process which addresses some of thelimitations of GLEAM Exgal, and present the GLEAMSGP images and source catalogue. The analysis of bothyears of GLEAM observations not only improves theimage sensitivity, but also provides a second epoch whichcan be used to study spectral variability on a one-yeartimescale.The layout of the paper is as follows. Section 2 out-lines the survey strategy employed in GLEAM obser-vations. Section 3 describes the improvements made tothe GLEAM calibration and imaging pipeline in thisrelease. Section 4 outlines the steps taken to constructand validate the GLEAM SGP catalogue. Our resultsare summarised in Section 5.Throughout this paper, we assume the conventionfor spectral index, α , where S ∝ ν α . Right ascensionis abbreviated as RA and declination is abbreviated asDec. The GLEAM year 1 and 2 observations were conductedwith Phase I of the MWA. Phase I of the MWA con-sisted of 128 16-crossed-pair-dipole tiles with baselinesextending to ≈ ≈ . δ + 26 . ◦ ) arcmin. Theprimary beam full width at half maximum (FWHM) at154 MHz is ≈ ◦ .The first year of GLEAM observations used forGLEAM Exgal were conducted between August 2013and June 2014. The whole sky south of Dec +30 ◦ wassurveyed in four week-long campaigns of meridian driftscans (Bernardi et al., 2013; Hurley-Walker et al., 2014)to obtain overlapping coverage in RA. Observations werecarried out at night to avoid contamination from theSun. The sky was divided into seven Dec strips ( − ◦ , − ◦ , − ◦ , − ◦ , − ◦ , +2 ◦ and +18 ◦ ) and one Decstrip was covered in a given night.Each night’s observing run was broken into a seriesof 2-minute scans in five frequency bands of bandwidth30.72 MHz centred at 87.7, 118.4, 154.2, 185.0 and215.7 MHz (hereafter 88, 118, 154, 185 and 216 MHz),cycling through the five frequency bands in 10 minutes.Frequencies between 134 and 139 MHz were avoided dueto Orbcomm satellite interference. A strong calibratorsource was observed in the five frequency bands at thebeginning of the observing run. The frequency and timeresolution of the correlator output were 40 kHz and 0.5 s, LEAM South Galactic Pole data release Table 1
GLEAM SGP observing parameters.
Date Year of RA Dec Hourobservation range (h) (deg) angle2013-08-10 1 21 − . −
27 02013-08-22 1 21 − . −
13 02013-08-25 1 21 − . −
40 02013-11-05 1 0 − −
13 02013-11-06 1 0 − −
40 02013-11-25 1 0 − −
27 02014-06-09 1 21 − −
27 02014-06-10 1 21 − −
40 02014-06-16 1 21 − −
13 02014-08-04 2 21 − −
27 -12014-08-05 2 21 − −
40 -12014-08-08 2 21 − −
13 -12014-09-15 2 21 − . −
27 +12014-09-16 2 21 − . −
40 +12014-09-19 2 21 − . −
13 +12014-10-27 2 21 . − −
27 -12014-10-28 2 21 . − −
40 -12014-10-31 2 21 . − −
13 -12014-12-17 2 1 − −
40 +12014-12-20 2 1 − −
13 +1respectively. More details on the observing strategy canbe found in Wayth et al. (2015).In the second year of GLEAM observations conductedbetween August 2014 and July 2015, twice the amountof observing time was spent surveying the same area ofsky at the same frequencies. The observing strategy wasadjusted as follows:(1) The observations were divided into eight week-longdrift scan campaigns, alternating between an hourangle of +1 and –1. This served to increase theeffective ( u, v ) coverage for each patch of the sky inthe final mosaic, and to observe some fields whenthe brightest, complex radio sources in the sky (theso-called ‘A-team’ sources) were below the horizonrather than in a sidelobe.(2) The frequency resolution of the correlator outputwas set to 10 kHz. The higher frequency resolutionwas chosen to increase the usefulness of the datasetfor spectral line and polarisation science. The timeresolution was set to 2 s to retain the overall datarate. A time resolution of 2 s does not lead tosignificant time-average smearing with the longestbaselines in the MWA Phase I array.
In this paper, we process a subset of the GLEAM year1 and 2 data at Decs − ◦ , − ◦ and − ◦ , and in theRA range 21 − h m − ◦ ). The SGP alsotransits through the MWA zenith ( ≈ − ◦ ), where theprimary beam has the highest sensitivity. Table 1 liststhe observations used for GLEAM SGP.We do not process the data from the lowest frequencyband (72–103 MHz) due to calibration errors associatedwith Fornax A and Pictor A entering the primary beamsidelobes. However, given that GLEAM Exgal is limitedby classical confusion in the lowest frequency band, nosignificant improvement in the sensitivity is expectedfrom combining the GLEAM year 1 and 2 data. The rmsnoise achieved in the lowest frequency band of GLEAMExgal is ≈
40 mJy/beam while Franzen et al. (2019) es-timate the classical confusion noise to be 30 mJy/beam.In the final source catalogue, we include flux densities ex-tracted from the GLEAM Exgal mosaics below 100 MHz,as explained in Section 4.The full GLEAM Exgal data reduction procedure isdescribed in detail in Hurley-Walker et al. (2017). In thissection, we summarise the main calibration and imagingsteps used in GLEAM Exgal, describe the changes madeto the data reduction process in GLEAM SGP, andcompare the sensitivity and dynamic range in the finalGLEAM Exgal and SGP mosaics.
The raw visibility data from each 2-min snapshot ob-servation were pre-processed using c o t t e r (Offringaet al., 2015): data affected by radio frequency interfer-ence (RFI) were flagged and the data were averaged toa time resolution of 4 s and a frequency resolution of40 KHz. For each night’s observing run, antenna am-plitude and phase solutions were derived for a sourcecalibrator observation at the beginning of the observ-ing run and applied to the entire night of drift scandata. The calibration was performed using the Com-mon Astronomy Software Applications (CASA ) task b a n d pa s s .The snapshot data were imaged and self-calibratedusing the w s c l e a n imager (Offringa et al., 2014),which corrects for wide-field w -term effects, and thefull-Jones m i t c h c a l algorithm developed for MWAcalibration (Offringa et al., 2016). At this stage, the30.72 MHz bandwidth of the data was divided intonarrower sub-bands of 7.68 MHz. Images of 7.68 MHzbandwidth at 20 frequencies distributed continuouslybetween 72 and 231 MHz, but avoiding 134–139 MHz,were created.The Molonglo Reference Catalogue (MRC; Large et al., http://casa.nrao.edu/ Franzen et al. f i t s _ wa r p (Hurley-Walker & Hancock,2018): the MRC and NRAO VLA Sky Survey (NVSS;Condon et al., 1998) catalogue were used to correctsource position offsets introduced in the snapshot imagesdue to ionospheric distortions.At each frequency, the snapshot images were correctedfor the primary beam using the analytical primary beammodel of Sutinjo et al. (2015) and mosaicked together.It was necessary to correct for residual Dec-dependenterrors in the flux scale due to errors in the adoptedprimary beam model. This was done by comparing themeasured flux densities of sources in the mosaic withtheir radio spectra as predicted by three catalogues: theVLA Low-Frequency Sky Survey redux (VLSSr; Laneet al., 2014) at 74 MHz, MRC at 408 MHz and NVSS at1.4 GHz. A map tracing the variation of the point spreadfunction (PSF) across the mosaic was generated usingsources known to be unresolved in higher resolution radiosurveys. The PSF map, which describes the apparentblurring of the PSF due to ionospheric smearing, wastaken into account when measuring source sizes andintegrated flux densities in the mosaics.The final image products consist of 20 Stokes I − ◦ ≤ Dec < . ◦ , the sensitivity is ≈
10 mJy/beam and the angularresolution ≈ . In a single night of GLEAM observing, roughly 10 TBof raw visibility data are generated. Using Dysco com-pression, Offringa (2016) showed that MWA data withtypical time and frequency resolutions used in processingcan be compressed by a factor of ≈ ≈ u, v )data are compressed using Dysco. We do not use the calibrator source observations tocalibrate the snapshot data. Instead, we rely on theGLEAM Exgal catalogue as a sky model to calibratethe data.For each GLEAM SGP snapshot, the sky model is constructed as follows. We select all sources in the cat-alogue which lie in the main lobe of the snapshot’sprimary beam (out to a radius of ≈ ◦ at 118 MHz and ≈ ◦ at 216 MHz). The catalogue contains spectral in-dices, α , and 200 MHz flux densities, S , derived frompower-law fits to the sub-band flux densities between76 and 227 MHz. Where possible, we use α and S toderive the integrated flux densities, S ν c , of the sources inthe model at the central frequency, ν c , of the snapshot.There is no measurement of the spectral index for 25 percent of the sources in the catalogue; the sources withmissing spectral indices are mostly the fainter ones forwhich there was insufficient signal-to-noise to make areliable measurement. If α and S are not available inthe catalogue, we derive S ν c by taking the sub-band fluxdensity closest in frequency to ν c and extrapolating it to ν c with α = − .
8, the typical spectral index of sourcesseen in GLEAM Exgal (Hurley-Walker et al., 2017).The morphology of the sources in GLEAM Exgal isbest characterised using the wide-band mosaic covering170–231 MHz as it has the best sensitivity and resolution.The GLEAM Exgal catalogue provides the major axis, a wide , minor axis, b wide , and position angle, θ wide , of eachsource, derived from Gaussian fitting in the wide-bandmosaic. The ratio of the integrated to peak flux density, R , in the wide-band mosaic can be used to distinguishbetween point-like and extended sources.We use our estimate of S ν c , and the source positionand α from the GLEAM Exgal catalogue, to characterisethe sources in the model. If α is not available in thecatalogue, we set the spectral index to –0.8. Sourcesin the sky model with R < . a wide , b wide and θ wide . Finally, anapparent flux density cut of 100 mJy is applied to limitthe processing time needed to generate the model visibil-ities, which is proportional to the number of componentsin the sky model. The final number of sources present inthe sky model varies between ≈ ,
000 at the lowest fre-quency (118 MHz) to ≈ ,
500 at the highest frequency(216 MHz).Calibration is performed using m i t c h c a l , scalingeach component in the sky model by a model of theprimary beam by Sokolowski et al. (2017). This pri-mary beam model is more accurate than that of Sutinjoet al. (2015) as every single dipole in the MWA tileis simulated separately, taking into account all mutualcoupling, ground screen and soil effects. As in GLEAMExgal, baselines shorter than 60 m are not used to per-form the calibration due to contamination from expectedlarge-scale Galactic emission.Hurley-Walker et al. (2017) removed areas within10 arcmin of the following A-team sources from theGLEAM Exgal catalogue due to the difficulty in cali-brating and imaging them at low frequencies: Centaurus
LEAM South Galactic Pole data release ≈
25 per cent of the snapshots are significantlyaffected. Most of these snapshots are located within ≈
20 deg of Fornax A and/or Pictor A, while a smallernumber of them are located in a sidelobe of Cygnus A.A different procedure is used to calibrate the datafrom these affected snapshots: we apply good calibrationsolutions derived from the GLEAM Exgal sky model forother snapshots at the same frequency and separatedby no more than 2 hours in time. We then performself-calibration following the same procedure as adoptedin GLEAM Exgal: a sky model is constructed fromthe observation itself and the model is used to furtherimprove the calibration solutions (see Hurley-Walkeret al., 2017 for details). For the remaining snapshots,this additional self-calibration loop is not performed asit is not found to significantly improve the calibrationsolutions.
Most of the processing time for GLEAM Exgal was spenton imaging the large number of snapshots within thesurvey. The GLEAM Exgal snapshots were imaged using w s c l e a n v1.10. The image size was set to 4000 × ≈ − . σ , was made by imaging the snapshot downto the first negative CLEAN component. The snapshotwas then re-imaged down to a CLEAN threshold of 3 σ .We image the ≈ w s c l e a n v2.5, which ismore efficient for large images thanks to the implemen-tation of the Clark CLEAN algorithm (Clark, 1980). Inminor CLEAN cycles, only the central portion of thesynthesised beam is used to subtract the CLEAN com-ponents from the image and only the largest CLEANcomponents are searched for. This is sufficient to findthe CLEAN components providing that the synthesisedbeam is well behaved. We use the same robust weightingof –1.0 but set the minimum ( u, v ) distance to 30 λ andtaper the weights with a Tukey transition of size 15 λ .This removes structure larger than ≈ . σ and simultaneously create amask containing all found components. We then con-tinue CLEANing with the constructed mask down to adeeper threshold of 1 σ . This CLEANing procedure is fully automated and allows stuctures to be CLEANeddown to the noise level around detected sources (Offringa& Smirnov, 2017).The flux scale of the snapshot images is expectedto be too low as a result of flux uncaptured in theGLEAM Exgal sky model (as mentioned above, allsources with S ν c , app <
100 mJy were removed fromthe model). In GLEAM Exgal, the MRC was used to seta basic flux scale for the snapshot images; a selected sam-ple of sources in the snapshot image were cross-matchedwith MRC and the measured flux densities were com-pared with those predicted from MRC. We adopt thesame procedure to correct the flux scale of the snapshotimages in GLEAM SGP except that we use the GLEAMExgal catalogue to set the flux scale.
We follow the exact same procedure as in GLEAM Exgalto combine the GLEAM year 1 and 2 snapshots at eachfrequency into mosaics and apply the Dec-dependentflux scale correction, with the following exceptions:(1) The synthesised beam size of the snapshots at eachfrequency is found to vary by up to ≈
10 per centdue to slight changes in the ( u, v ) coverage of theobservations. Before mosaicking, each snapshot isconvolved with a Gaussian to obtain an identicalsynthesised beam at each frequency.(2) In order to improve the accuracy of the flux scale,we use the primary beam model by Sokolowski et al.(2017) in the mosaicking step.(3) The mosaicking process results in a set of 16 imagesbetween 107 and 227 MHz, each with a bandwidthof 7.68 MHz. In GLEAM Exgal, the most sensitivecombined image was obtained by combining theeight highest frequency sub-band mosaics at 170–231 MHz. In GLEAM SGP, we find that combiningthe four highest frequency sub-band mosaics at 200–231 MHz results in a better compromise betweensensitivity and resolution.As mentioned in Section 3.1, ionospheric perturbationscan cause sources to be slightly smeared out in themosaicked images. We therefore generate maps of thespatial variation of the PSF across each of the mosaicsusing the method adopted in GLEAM Exgal. The mean ± standard deviation of the major and minor axes of thePSF in the wide-band mosaic are (2 . ± .
05) arcminand (1 . ± .
05) arcmin, respectively.The wide-band mosaic centred at 216 MHz is shownin Fig. 1. We use this image for source detection, asdescribed in Section 4. F r a n z e n e t a l . Figure 1.
The GLEAM SGP wide-band mosaic centred at 216 MHz. The grey-scale is linear and runs from –50 to 100 mJy/beam. The blue line indicates the catalogue boundary ofthe mosaic, chosen as described in Section 3.3. The black cross marks the SGP.
LEAM South Galactic Pole data release We compare the rms noise in the GLEAM Exgal andSGP wide-band mosaics. We create rms noise maps of thetwo mosaics using
B A N E (Hancock et al., 2018). Usingthese noise maps, we define an area of sky where the rmsnoise is lower in GLEAM SGP: 20 h m < RA < h m and − ◦ < Dec < − ◦ . In this 5,113 deg area ofsky, hereafter referred to as the GLEAM SGP region,the mean rms noise in GLEAM SGP (4.7 mJy/beam)is ≈
40 per cent lower than that in GLEAM Exgal(7.6 mJy/beam). The best improvement in the rms inGLEAM SGP is by a factor of ≈ ≈ √ u, v )coverage and improved deconvolution of the snapshotimages. Since sidelobe confusion is the dominant noisecontribution in GLEAM Exgal above ≈
100 MHz, mostof the reduction in the rms noise in GLEAM SGP resultsfrom the lower sidelobe confusion noise.Fig. 2 compares an example 25 deg of sky in theGLEAM Exgal and SGP wide-band mosaics. The re-duction in image artefacts around bright sources in theGLEAM SGP mosaic likely results from the improvedcalibration and deconvolution of the snapshots, as wellas the larger number of snapshots contributing to anypatch of sky. The diffuse Galactic synchrotron emissionvisible in the GLEAM Exgal mosaic is also largely re-moved in the GLEAM SGP mosaic as a result of the( u, v ) taper applied to the GLEAM SGP data. We create additional mosaics for the GLEAM year 1and 2 data by combining the GLEAM year 1 and 2snapshots at each frequency into mosaics separately. TheGLEAM year 1 data were taken almost entirely over theperiod August–November 2013 and the GLEAM year2 data were taken over the period August–December2014, as shown in Table 1. These two epochs of datatherefore provide a unique opportunity to search forlow-frequency variability in the flux density over a largefractional bandwidth on a timescale of approximatelyone year. The mean rms noise in the GLEAM year 1 and2 wide-band (200–231 MHz) mosaics within the GLEAMSGP region is 6.5 and 5.5 mJy/beam, respectively.
Figure 2.
Example 25 deg of sky containing several brightsources from the GLEAM Exgal wide-band mosaic at 170–231 MHz(top) and the GLEAM SGP wide-band mosaic at 200–231 MHz(bottom), highlighting the improvement in the rms in GLEAMSGP. The grey scale is linear and runs from –20 to 50 mJy/beamin both panels. Franzen et al.
Table 2
Comparison of the GLEAM Exgal and SGP survey properties in the GLEAM SGP region (20 h m < RA < h m and − ◦ < Dec < − ◦ ). Values are given as the mean ± the standard deviation. The statistics shown are derived from theGLEAM Exgal wide-band mosaic at 170–231 MHz and the GLEAM SGP wide-band mosaic at 200–231 MHz. The RA andDec astrometric offsets show the degree to which the source positions agree with NVSS and SUMSS; the RA offset is givenby RA NVSS / SUMSS − RA GLEAM and the Dec offset by Dec
NVSS / SUMSS − Dec
GLEAM . The external flux scale error applies toall frequencies and shows the degree to which the source flux densities agree with other published surveys. The internal fluxscale error also applies to all frequencies and shows the internal consistency of the flux scale.
Property GLEAM Exgal GLEAM SGPNumber of sources 85,981 108,851RA astrometric offset (arcsec) − . ± . − . ± . − . ± . − . ± . . ± . . ± . ± ± ± ± Following Hurley-Walker et al. (2017), we perform blindsource finding on the wide-band mosaic covering 200–231 MHz to obtain a reference catalogue. We then extractthe flux densities of each source within the referencecatalogue in the sub-band images.The source finding on the wide-band mosaic is per-formed as follows. We first use b a n e to estimate thebackground emission and rms noise across the mosaic.Next, we run a e g e a n (Hancock et al., 2012, 2018)on the mosaic using a detection threshold of 5 × thelocal rms. The spatial variation of the PSF is takeninto account using the PSF map. Each detected sourceis characterised by a e g e a n as an elliptical Gaussiancomponent. Six parameters are fitted for each compo-nent: the peak RA and Dec, peak flux density, majorand minor axes, and position angle.In order to remove potential spurious detections, allsources within 0.5 deg of Fornax A are discarded. Af-ter removing these sources, the total number of sourcesdetected in the GLEAM SGP region is 108,851. In com-parison, the GLEAM Exgal catalogue contains 85,981sources in the GLEAM SGP region.For each source in the reference catalogue, we extractthe flux density in each of the 16 sub-band images be-tween 107 and 227 MHz using the ‘priorised fitting’ modeof a e g e a n . The expected shape of the source in thesub-band image is derived by a e g e a n given its shapein the wide-band image, and the local PSFs from thewide-band and sub-band images. A fit is performed forthe peak flux density of each source; the position andnewly-determined shape of the source are not allowedto vary.We use the four lowest frequency sub-band mosaicsfrom GLEAM Exgal to extract additional flux densitymeasurements between 76 and 99 MHz for all sources within the reference catalogue, via priorised fitting.The advantage of this priorised fitting approach is thatit provides measurements for all sources in the referencecatalogue across the full frequency range without havingto rely on position-based cross-matching of catalogues.The positions and morphologies of the sources are mostprecisely determined in the wide-band image which hasthe best resolution and sensitivity, and this informationis used to constrain the source flux densities in the sub-band images. Since the flux densities in the sub-bandimages are not extracted via blind source finding, nosignal-to-noise ratio (SNR) threshold is applied to theflux densities from the sub-band images. Sources maytherefore be detected well below 5 σ in the sub-bandimages, or even have negative flux densities. We calculate the spectral indices between 76 and227 MHz of the GLEAM SGP sources from the 20 sub-band flux densities. For the spectral index of a sourceto be calculated, it must have a positive flux densitymeasurement in each of the 20 sub-bands; this is thecase for 77 per cent of the sources in the catalogue. Forthese 83,328 sources, we calculate α using a weightedleast-squares approach. The flux density error in eachsub-band is taken as the sum in quadrature of the Gaus-sian fitting error (as calculated by a e g e a n ) and acalibration error of 2 per cent. We estimate the internalflux calibration error to be 2 per cent from the reduced χ statistic for bright sources: at high flux densities( S
216 MHz (cid:38) χ value shouldbe close to 1.0 if the internal flux calibration error hasbeen well estimated. We find this to be the case whenusing an internal flux calibration error of 2 per cent.In addition to the spectral index, the fitted 200-MHzflux density and reduced χ value from the least-squaresfitting are provided in the catalogue. The reduced χ LEAM South Galactic Pole data release N u m b e r Figure 3.
The spectral index distribution of the GLEAM SGPsources measured using the 20 sub-band flux densities between 76and 227 MHz. Only sources with reduced χ < .
93 and δα < . value can be used to assess the quality of the fittedspectral index and 200-MHz flux density: for 18 degreesof freedom, P (reduced χ > . <
1% and P (reduced χ > . < . χ < .
93 and spec-tral index error, δα < .
5. Fig. 3 shows the spectralindex distribution of these sources. The mean and me-dian spectral indices are –0.81 and –0.82, respectively.A tiny fraction of the sources with reduced χ < . δα < . α < − ≈ ≈
10 times higher in thewide-band image, its measured spectral index may besignificantly affected by confusion.
A common method for classifying sources as point-like orextended is through the ratio of their integrated to peakflux densities. Fig. 4 shows the ratio of the integratedto peak flux density in the wide-band image, SS p , asa function of the SNR, for all GLEAM SGP sources.Instances where SS p < . I n t eg r a t ed / pea k f l u x den s i t y Figure 4.
The ratio of the integrated to peak flux density as afunction of the SNR for all GLEAM SGP sources detected in thewide-band image. Sources classified as point-like are shown in redand as extended in blue. uncertainties into account: SS p > . a s(cid:18) σ local S p (cid:19) + (cid:15) , (1)where σ local is the local rms noise and (cid:15) the internal fluxscale error. We set (cid:15) = 0 .
02 based on the analysis carriedout in Section 4.1. Following a similar approach to thatof Butler et al. (2018), we set the value of a such that95 per cent of the sources with SS p < . SS p = 1 . − a s(cid:18) σ local S p (cid:19) + (cid:15) . (2)The resulting value of a is 2.77. At high SNR wherecalibration errors dominate, SS p > .
06 for a source tobe classified as extended; at an SNR of 5, SS p > .
56 fora source to be classified as extended.Using Equation 1, 9.2 per cent of the GLEAM SGPsources are classified as extended, where the beam sizeis ≈ Simulations in the image plane are used to quantifythe completeness of the source catalogue. Following thesame method as in GLEAM Exgal, we inject artificialpoint sources with flux densities ranging between 150
Franzen et al.
20 50 100 300 S sim [ mJy ] C o m p l e t e n e ss [ % ] Figure 5.
Estimated completeness across the GLEAM SGP regionas a function of S
216 MHz . The black curve shows the mediancompleteness and the shaded area the 10–90 percentile range. and 300 mJy into the wide-band mosaic used for sourcedetection. We then create maps tracing the variationof the completeness across the sky at the various fluxdensity levels. The completeness at any pixel positionis given by the fraction of simulated sources that aredetected above 5 σ in a circle of radius 6 ◦ centred onthe pixel. Full details of the procedure are explained inHurley-Walker et al. (2017). The completeness maps in f i t s format can be obtained from the supplementarymaterial.The black curve in Fig. 5 shows the median com-pleteness across the GLEAM SGP region as a functionof S
216 MHz ; the shaded area indicates the 10–90 per-centile range. The completeness of the source catalogueis estimated to be 50% at ≈
25 mJy, rising to 90% at ≈
50 mJy.In order to assess the reliability of the source cata-logue, we use
A e g e a n to only search for sources withflux densities below − σ in the wide-band image. Intotal, 7 sources with negative peaks below − σ are de-tected in the GLEAM SGP region. Assuming the noisedistribution is close to symmetric about zero, we canexpect to find an approximately equal number of falsepositive sources in the same area. The total numberof sources detected above 5 σ is 108,851. We thereforeestimate the catalogue reliability to be:1 . − ,
851 = 99 . . (3) The positions of the GLEAM SGP sources are ex-tracted from the wide-band mosaic at 200–231 MHz,used for source detection. In order to verify the as- trometry, we cross-match our catalogue with the higher-resolution NVSS catalogue at δ ≥ − . ◦ and the higher-resolution Sydney University Molonglo Sky Survey(SUMSS; Mauch et al., 2003) catalogue at δ < − . ◦ ;NVSS has an angular resolution of 45 arcsec and SUMSShas an angular resolution of 45 × cosec | δ | . We onlyinclude unresolved ( S int /S pk < . NVSS / SUMSS − RA GLEAM , and Dec offset, ∆Dec = Dec
NVSS / SUMSS − Dec
GLEAM , for the 10,741 unresolved, isolated sourcesin common between GLEAM SGP and NVSS/SUMSS.Sources with GLEAM SGP signal-to-noise ratios (SNRs) ≥ σ RA , cal and σ Dec , cal in RA andDec. The rms deviation of ∆RA is 3.3 arcsec and therms deviation of ∆Dec is 3.1 arcsec. We therefore set σ RA , cal = 3 . σ Dec , cal = 3 . σ RA and σ Dec in RA and Dec to σ RA = q σ , cal + σ , fit (4) σ Dec = q σ , cal + σ , fit , (5)where σ RA , fit and σ Dec , fit are the Gaussian fitting errorscalculated by a e g e a n , which account for the imagenoise. As part of the mosaicking procedure, Dec-dependentflux scale corrections were applied by comparing the fluxdensities of sources in the mosaics with the flux densitiespredicted from the SEDs formed from VLSSr, MRCand NVSS flux density measurements. We calculatethe external flux scale error at each frequency as thestandard deviation of a Gaussian fit to the remainingvariation in the ratio of the predicted to measured sourceflux densities. The percentage uncertainties lie between6.6 and 7.9 per cent. For simplicity, we set the externalflux scale error to 8 per cent at all frequencies.Hale et al. (2019) produced an image of the XMMLarge-Scale Structure (XMM-LSS) field with LOFAR at144 MHz. Their image is centred at RA = 02 h m , Dec The results obtained when cross-matching GLEAM SGPsources with NVSS and SUMSS separately are similar: when cross-matching with NVSS, the rms deviation of ∆RA is 3.2 arcsec andthe rms deviation of ∆Dec is 3.2 arcsec. When cross-matchingwith SUMSS, the rms deviation of ∆RA is 3.6 arcsec and the rmsdeviation of ∆Dec is 2.7 arcsec.
LEAM South Galactic Pole data release
30 20 10 0 10 20 30RA (arcsec)3020100102030 D e c ( a r c s e c ) Figure 6.
RA and Dec offsets for GLEAM SGP sources cross-matched with NVSS or SUMSS, as described in the text. Sourceswith GLEAM SGP SNRs > 100 are shown in red and the rest ofthe sources in black. = − ◦ and covers an area of ≈
27 deg . The centralrms noise is 280 µ Jy/beam and the angular resolution 7.5by 8.5 arcsec. Most of the area covered by this image lieswithin the GLEAM SGP region (the northern extremityof the image at Dec > − ◦ lies outside the GLEAMSGP region).Only relatively bright ( S
216 MHz (cid:38) χ < .
93) and spectral indexerrors greater than 0.5. This leaves a total of 82 sourcesto use for the flux density comparison.In Fig. 7, the ratio, R , of the GLEAM SGP to LOFARflux density is plotted as a function of the GLEAM SGP GLEAM SGP flux density at 216 MHz (mJy) G L E A M S G P / L O F A R f l u x d e n s i t y a t M H z ( m J y ) Figure 7.
Ratio of the GLEAM SGP to LOFAR flux density as afunction of the LOFAR flux density for a sample of 82 sources inthe XMM-LSS field. The GLEAM SGP flux densities are measuredfrom the wide-band image at 216 MHz. The LOFAR flux densitiesoriginate from Hale et al. (2019). The dashed horizontal lineindicates equal flux density values. The red horizontal line marksthe median flux density ratio. flux density. The GLEAM SGP flux densities are onaverage consistent with the LOFAR flux densities at the ≈ R is 0 . ± . . ± .
03. The relatively largescatter in R (the standard deviation of R is 0.23) is likelydue to the large difference in sensitivity and resolutionof the two catalogues, as well as the inclusion of sourceswith SNRs as low as ≈ R ( > .
5) all havelow SNRs in GLEAM SGP ranging between 6.0 and 10.8.Their GLEAM SGP flux densities may be biased highdue to the Eddington bias (Eddington, 1913) close tothe survey detection limit. Another potential cause ofthe discrepancy is missing extended flux in the LOFARimage.
The GLEAM SGP catalogue gives the position of eachsource in the wide-band image, and the integrated fluxdensity and shape of each source in the wide-band andsub-band images. The local PSF at the location of eachsource is provided at each frequency. The cataloguecontains 108,851 rows and 314 columns. Columns 1–311are exactly the same as those in the GLEAM Exgalcatalogue except that the measurements derived fromthe wide-band image (columns 10–26 with the subscript‘wide’) are centred at 216 MHz rather than at 200 MHz.A description of the 311 columns in the GLEAM Exgalcatalogue can be found in Appendix A of Hurley-Walkeret al. (2017). The remaining columns are defined as2
Franzen et al. follows:
Columns 312 and 313 – Best estimate of the 200 MHzintegrated flux density, S , and its associated error, σ S , in Jy. S and σ S are set to the fitted 200 MHzflux density and its error, providing the spectrum is wellfit by a power law (reduced χ < . S isestimated by extrapolating the 216 MHz integrated fluxdensity from the wide-band image assuming α = − . σ S is set to the scaled uncertainty on the wide-bandflux density. These two columns are provided in order toensure that all sources in the GLEAM SGP cataloguehave a measurement of their 200 MHz flux density, as isthe case in the GLEAM Exgal catalogue. Column 314 – Extended flag: point-like (0) or extended(1) (see Section 4.2).The electronic version of the GLEAM SGP catalogueis available from VizieR.
This work presents images and an extragalactic sourcecatalogue from combining both years of GLEAM ob-servations at 72–231 MHz conducted with Phase I ofthe MWA. The data release covers a 5,113 deg areaof sky centred on the SGP at 20 h m < RA < h m and − ◦ < Dec < − ◦ . The typical rms noise level is ≈ ≈ ≈
40 per cent lowerthan in GLEAM Exgal, which is solely based on the firstyear of GLEAM observations, as a result of the longerintegration times, better ( u, v ) coverage and improvedprocessing. A total of 108,851 source components aredetected above 5 σ at 216 MHz and the source catalogueincludes 72–231 MHz spectral indices for 77 per cent ofthe components. The catalogue is estimated to have acompleteness of 50 per cent at 25 mJy and a reliabilityof 99.99 per cent.The GLEAM SGP data reduction procedure largelyfollows that of GLEAM Exgal but we make significantimprovements in a number of areas: we use the GLEAMExgal catalogue as a sky model to calibrate the snap-shot data. The burden on computational resources interms of storage space and processing time is greatlyreduced thanks to the use of Dysco compression and theimplementation of the Clark CLEAN algorithm in w s -c l e a n . We use a new full embedded element primarybeam model (Sokolowski et al., 2017) in the calibrationand mosaicking to improve the accuracy of the flux scaleacross the mosaics.The GLEAM SGP data are well suited to large-scale studies of extragalactic source populations. Rosset al. (2021) have searched for variable sources, includ-ing blazars and compact-steep-spectrum (CSS) sources,by comparing the flux densities and spectral shapes of sources in the GLEAM SGP year 1 and 2 mosaics. Theirstudy represents the largest survey of low-frequency spec-tral variability to date, using quasi-simultaneous fluxdensity measurements over a large fractional bandwidth.Franzen et al., in preparation, are using the GLEAMSGP data in combination with optical spectroscopy fromthe 6dF Galaxy Survey (6dFGS; Jones et al., 2009) todetermine the local radio luminosity function for AGNand star-forming galaxies at 200 MHz, and characterisethe typical radio spectra of these two populations.In 2018, the MWA was upgraded with the addition ofa further 128 tiles, 56 of which were deployed on longbaselines, doubling the maximum baseline of the array(Wayth et al., 2018). In the MWA Phase II configuration,the angular resolution at 154 MHz is ≈ . https://wiki.mwatelescope.org/ . None.
This scientific work makes use of the Murchison Radio-astronomy Observatory, operated by CSIRO. We acknowl-edge the Wajarri Yamatji people as the traditional ownersof the Observatory site. Support for the operation of theMWA is provided by the Australian Government (NCRIS),under a contract to Curtin University administered by As-tronomy Australia Limited. We thank the anonymous refereefor helpful comments, which have substantially improved thispaper. We acknowledge the Pawsey Supercomputing Centrewhich is supported by the Western Australian and AustralianGovernments.
REFERENCES
Beardsley A. P., et al., 2016, ApJ, 833, 102Beardsley A. P., et al., 2019, PASA, 36, e050Bernardi G., et al., 2013, ApJ, 771, 105Briggs D. S., 1995, in American Astronomical SocietyMeeting Abstracts. p. 112.02Butler A., et al., 2018, A&A, 620, A3Callingham J. R., et al., 2017, ApJ, 836, 174Clark B. G., 1980, A&A, 89, 377Condon J. J., Cotton W. D., Greisen E. W., Yin Q. F.,Perley R. A., Taylor G. B., Broderick J. J., 1998, AJ,115, 1693